OBJECTIVE: Improvement in the early detection of primary and metastatic lung cancer using radiologic techniques. APPROACH: First, the improvement of film reader performance to decrease the number of misses and, second, improving the image using image subtraction techniques to make earlier or smaller masses visible. The work is divided into theoretical studies (to provide a data base and working hypothesis) and practical studies (to actually improve detection). The theoretical studies include: (1) the collection of data about the properties of lesions and their surrounding structures that were missed by competent film readers; (2) a study of the relation of the surrounding structures, called structured noise, to the probability of detection of lesions; and (3) an analysis of the tolerance of image registration required to detect lung nodules by the subtraction of serial chest films. The practical studies include: (1) the use of visual feedback to assist visual search; (2) the use of optical image processing to subtract structural noise from chest films; and (3) the use of digital subtraction of fluoroscopic images obtained during one heartbeat to detect small changes in vessel pulsation that are indicative of the presence of a primary lung tumor.